🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

We have liftoff! After a successful launch this weekend, the first FireSat satellite is now orbiting Earth 🛰️ It’s the first of a 50+ satellite constellation that will help detect + track wildfires as small as 5x5 meters, using AI. Huge thanks to partners Muon Space Earth Fire Alliance

🧵Wildfires are a growing threat globally — but existing tech just isn't very good at detecting and tracking them. Here's how we worked with Muon Space Earth Fire Alliance Moore Foundation to launch FireSat, a constellation of satellites to spot fires when they're still small ⬇️

love me a good pareto frontier...

All the evidence is that Gemini 2.5 Pro represents a new level of LLM that is much more capable than any released model, including other big reasoners. It got 24.4% on the Math Olympiad test that every other model bombed at, for example. I am very curious to see the system card.

This was actually very impressive. An assignment in my class is to determine the price & TAM of a product that can give the calorie count of food using a new form of spectroscopy, with $75 COGS. The new Gemini 2.5 Deep Research got all of that & much more. It used to take hours.

Ilya Sutskever's Safe Superintelligence will be using Google's TPUs.

Beautiful paper. Scientists solved the structure of myosin—the protein that forms contractile filaments in muscle cells—in multiple configurations using Cryo-EM. They captured the entire power stroke action (where myosin pulls on actin to shorten the muscle) at4.4 Å resolution.

Amazing project, congrats all!

excited to talk to folks at berkeley about brains (and more) -- stop by if you're around tonight

In collaboration with HHMI Janelia & Harvard, we introduce ZAPBench, a whole-brain activity dataset and benchmark with single cell resolution of a larval zebrafish, which will enable development and comparison of more accurate brain activity models. →goo.gle/44rEpsn

Google Research, Harvard University and HHMI | Janelia have created the Zebrafish Activity Prediction Benchmark (ZAPBench), which can help researchers more accurately model brain activity in larval zebrafish — another step toward understanding more complex brains. goo.gle/42LMMME

Can we predict future brain activity in a small vertebrate? Stop by the Google booth at 12PM ICLR 2026, for the introduction of ZAPBench: a benchmark to forecast activity in a whole larval zebrafish brain with 70k+ neurons. openreview.net/forum?id=oCHsD…

late post, but heartfelt thanks to Viren Jain for coming through for our second to last BioML @ Berkeley seminar for the semester! standing room only and back to back questions until we called it... I'm thinking we might need a bigger venue 🫥

Today, in collaboration w/ colleagues at the Institute of Science and Technology Austria (ISTA), we report the first-ever method for using light microscopy to comprehensively map all the neurons & their connections in a block of mouse brain tissue. More →goo.gle/4maUNUs

This was another great paper to work on. EM has been the standard for connectomic research since forever, but is a pain if you want to want to label molecules or cells. This paper makes an important step towards using light microscopy for connectomics. nature.com/articles/s4158…

A fluorescence-microscopy method for tracing neuronal connections in the brain could make connectomics studies more widely accessible for neuroscientists go.nature.com/3GMsATv

Nature research paper: Light-microscopy-based connectomic reconstruction of mammalian brain tissue go.nature.com/42MnD5S

An important addition to the Nature paper, beyond what was shown in the 2024 preprint, is 12 rounds of iterative imaging and sectioning of a LICONN volume, achieving 205 microns in axial extent (native scale) with manual tracing of axons: nature.com/articles/s4158…

Wouldn't it be great if we could not only image large connectomic volumes, but also completely reconstruct them? And if a whole mouse brain project didn't cost billions? With the PATHFINDER preprint (biorxiv.org/content/10.110…), we preview a future where it doesn't have to.

Incredible progress from Michał Januszewski on further automating connectomics, leading to order of magnitude reductions in error rates and estimated proofreading costs.